Although increasing numbers of young people are trying to quit e-cigarettes, personalized tools to support vaping cessation remain limited. We aimed to build a machine learning model to predict individual probability of short-term relapses and identify person-specific barriers to successful cessation. Data were taken from the “Stop Vaping Challenge” smartphone app. We included past 30-day e-cigarette users aged 15–35 years (n = 311) who completed 387 quit challenges. Feature selection minimized number of predictors while maximizing predictive ability. We built multiple GBM survival models with different sets of predictors to predict time to vaping relapse. The five-feature model yielded the best performance (C-index 0.751), thereby was selected as the final model. These five features were: self-confidence in quitting, intention to quit, average e-liquid used per week, time to first vape and mood trend during challenge. We stratified the challenges by the individual relapse risk by 7 days into low-, medium-, and high probability of quit success. This approach can inform tailored quit plans for vaping cessation. SHAP analysis demonstrated individual-level barriers to cessation, which can guide the development of personalized, predictor-centric, adaptive behavioral interventions. However, future research is needed to implement the model in real-world settings and evaluate its effectiveness and generalizability.
Kundu et al. (Sat,) studied this question.